Most teams optimising a landing page are staring at the wrong number. Bounce rate has become the default health metric for landing pages, and it's a genuinely poor one - not because it's meaningless, but because it collapses too many different visitor behaviours into a single figure that can't tell you what to fix. A visitor who reads every word of your page and then closes the tab looks identical to one who arrived, glanced at the headline, and left in under three seconds. Same bounce. Completely different problem.
The metrics that actually diagnose landing page performance are behavioural, not aggregate — and they point to specific things you can change. They tell you where visitors stop reading, which CTAs they ignore, and whether the people who do convert are the right people.
The metrics that actually tell you something
Scroll depth is the first thing to add if you haven't already. It tells you what percentage of visitors reach specific points on the page - typically measured at 25%, 50%, 75%, and 100%. A page where 80% of visitors never pass the fold has a hero problem. A page where visitors scroll to 75% and then drop off has a closing problem - the copy is holding attention but failing to convert it. Time on page has the same problem.
Time on page is useful but needs context. A high average time on page sounds positive until you realise it might mean visitors are confused, re-reading sections trying to understand what the product actually does. Pair it with scroll depth and you get a clearer picture: long time on page combined with low scroll depth often signals that visitors are stuck near the top, not engaged throughout. Short time on page with high scroll depth, on the other hand, can indicate a fast, confident reader who found what they needed - or didn't.
CTA click rate is the metric most teams underweight. Not conversion rate - click rate on the primary CTA specifically. If your page has a 2% conversion rate but a 12% CTA click rate, the problem is downstream: your sign-up flow, your onboarding, your pricing page. If your CTA click rate is 1.5%, the problem is the page itself — and that distinction determines whether you fix the funnel or rewrite the hero. Ferguson's audit data shows that 55% of pages fail the single primary CTA check - meaning more than half the pages audited are splitting visitor attention across multiple competing actions, which makes CTA click rate almost impossible to interpret cleanly.
What session recordings reveal that analytics can't
A session recording tool - one that captures mouse movement, clicks, and scroll behaviour as video replays - surfaces things that aggregate metrics bury. Watch ten sessions of visitors who arrived from your highest-intent traffic source and didn't convert. You'll almost always find one of three patterns: they're clicking on something that isn't a link (a visual element they expected to be interactive), they're hovering over your pricing section for a long time before leaving (a trust or clarity problem), or they're scrolling past your CTA without pausing (a visibility or relevance problem).
The hover-on-pricing pattern is particularly telling. Visitors who linger on pricing before leaving are unconvinced, not uninterested — the fix is usually more specificity about what's included, clearer anchoring between tiers, or better objection handling near the price point. Across Ferguson's corpus of 155 audited pages, objection coverage fails on 82% of pages - the single most common failure in the dataset. If visitors are lingering on pricing and leaving, that number explains a lot.
Heatmaps complement recordings by showing aggregate click patterns. The thing to look for isn't where people are clicking - it's where they're clicking that you didn't expect. Navigation links that pull visitors off the page. Images they're trying to click through. Testimonial names they're clicking, presumably to verify the person is real. If visitors are clicking testimonial names to verify they're real, your social proof isn't doing its job.
Segmenting by traffic source changes the diagnosis entirely
Aggregate conversion rate is almost useless for diagnosis. A page converting at 3.5% overall might be converting paid search traffic at 7% and organic traffic at 1.2% - and those two populations need completely different things from the page. Paid search visitors often arrive with high intent and specific expectations set by the ad copy. If the landing page doesn't immediately match what the ad promised, they leave fast. Organic visitors may be earlier in their research, less certain about what they need, and more likely to respond to educational framing and social proof.
Segment your conversion rate, scroll depth, and CTA click rate by at least three sources: paid, organic, and direct. If you have referral traffic from a specific partner or publication, segment that separately too - it often converts at a very different rate because the audience arrives pre-warmed. A page that looks mediocre overall might be performing well for one source and catastrophically for another.
Device type is the other segmentation that consistently reveals problems. Mobile visitors on a page optimised for desktop often show dramatically lower scroll depth and CTA click rates - not because they're less interested, but because the page is harder to navigate, the CTA button is awkwardly placed, or the hero section takes up the entire screen and gives no reason to scroll. This is worth checking before you invest time in copy changes.
The post-conversion metrics most teams ignore
Landing page performance doesn't end at the conversion event. If your page is generating trial sign-ups but those trials aren't activating, the page may be attracting the wrong visitors or setting the wrong expectations. Lead quality is a landing page metric — you just have to follow conversions into your funnel to see it.
One practical way to track this: tag conversions by the landing page variant or traffic source they came from, then follow those cohorts through your funnel. If visitors from one source convert on the page at a high rate but churn within two weeks, the page is probably over-promising or under-qualifying. If visitors from another source convert at a lower rate but retain well, the page might actually be doing a better job of setting accurate expectations - and the lower conversion rate is a feature, not a bug. If you haven't tagged conversions by source and checked retention, you don't yet know whether your page is working.
Post-CTA clarity is a related issue that shows up in Ferguson's audit data: 63% of pages fail this check, meaning visitors who click the CTA often land somewhere that doesn't match what they expected. That mismatch shows up as drop-off in your sign-up flow, not on the landing page itself - which is exactly why teams miss it when they're only watching landing page metrics. If you're seeing high CTA click rates but low completed sign-ups, landing page copy mistakes around expectation-setting are often the culprit.
Building a measurement stack that's actually usable
The temptation is to instrument everything and end up with a dashboard nobody looks at. A minimal stack — one you'll actually use — covers five things:
- Scroll depth at 50% and 75% - tells you whether the page is holding attention past the fold
- CTA click rate on your primary action - segmented by traffic source, not just overall
- Conversion rate by device type - catches mobile problems before they compound
- A session recording review of 10 to 15 non-converting sessions per week from your highest-intent source - qualitative, but often the fastest path to a real insight
- Lead quality signal from downstream - even a rough one, like trial activation rate tagged by source
If you want a structured starting point before you dig into behavioural data, a landing page audit checklist can surface the structural issues that metrics alone won't catch - things like whether your headline passes a five-second comprehension test or whether your social proof is positioned where it can actually do work. Ferguson's audit data shows that 41% of pages fail the five-second test, which means nearly half the pages in the dataset are losing visitors before any of your carefully tracked metrics even have a chance to register.
The Ferguson benchmark report has more on how pages in the audit corpus compare across these dimensions if you want a reference point for where your numbers sit relative to others.